Abstract:
Existing global registration methods are prominently iterative. They require iterations and can be sensitive to point densities and noise. In contrast, closed-form soluti...Show MoreMetadata
Abstract:
Existing global registration methods are prominently iterative. They require iterations and can be sensitive to point densities and noise. In contrast, closed-form solutions provide a more robust estimation model and do not involve iterations. In this letter, we present a global closed-form refinement for the terrestrial laser scanner (TLS) data registration problem. Our proposed method segments the task in three key steps. First, the method exploits a plane-based approach to compute the transformation parameters, resulting in the pairwise registration between point clouds. Second, we place all rotation parameters into a common coordinate system exploring one operation of the quaternions' properties. Third, we constrain the refined rotation parameters to globally refine the translation. The effectiveness of the proposed method is demonstrated with a TLS data set. Experiments have demonstrated that the proposed method can properly create a consistent 3-D map of outdoor environments with accuracy at the decimeter level.
Published in: IEEE Geoscience and Remote Sensing Letters ( Volume: 14, Issue: 7, July 2017)